I assume that: developers can create a micro-app with a dead simple API ... users run and discover these micro-apps from within the human.io mobile application.
An example micro-app would be where I create a bookmarklet that grabs an image, it sends it to human.io with the text "please identify this shirt" and then a human.io user can choose to run and solve that "micro-app"
It's confusing because micro-app implies repetition, whereas they are more like tasks: they have a "completion." This is a more powerful, easier to use, mechanical turk where the turks are on mobile.
We were hoping for a quiet soft-launch so we could get more API feedback before a real launch (including use-cases like you suggested.)
The events are pushed back to a webserver, and you decide what to do there. If you already have an application and data storage etc already, this is basically the easiest thing to do. We felt this flexibility would lead to the most interesting initial apps; additionally, the vast majority of early adopters would probably not love being sandboxed.
 we also have a hanging HTTP api; call /next_event and it hangs until it has events for you. this lets you talk to the system without having a public open port.